一种新的降维算法PCA_LLE研究及其在图像识别中的应用<sub><sup> </sup></sub>
A New Dimensionality Reduction Algorithm PCA_LLE and Its Application in Image Recognition
投稿时间:2019-07-08  修订日期:2019-07-08
DOI:
中文关键词: 降维  主成分分析  局部线性嵌入  图像识别  PCA_LLE
英文关键词: dimensionality reduction  PCA  LLE  image recognition  PCA_LLE
基金项目:国家自然科学基金项目(面上项目,重点项目,重大项目)
作者单位E-mail
蓝雯飞 中南民族大学 计算机科学学院 lanwenfei1@163.com 
汪敦志 中南民族大学 计算机科学学院  
张盛兰 中南民族大学 计算机科学学院  
摘要点击次数: 301
全文下载次数: 0
中文摘要:
      图像等高维数据在利用传统分类方法进行分类时会面临“维度灾难”,通过降维将原始高维数据映射到低维空间,是缓解维度灾难的重要途径。线性降维方法PCA考虑的是数据的全局结构;流形学习降维方法LLE考虑数据的局部结构。本文通过结合PCA与LLE两种降维方法,提出新的PCA_LLE算法,使它们优势互补。在手写体数字数据集上进行实验,先对数据集降维,再用K近邻算法对降维后的数据分类。实验结果表明融合两种算法的PCA_LLE降维方法较原来的PCA和LLE算法准确率均有了提升。而且新算法PCA_LLE对新样本的降维时间较LLE算法减少很多。
英文摘要:
      High-dimensional data such as image faces a “dimension disaster” when it is classified using traditional classification methods. Mapping the original high-dimensional data to low-dimensional space through dimensionality reduction is an important way to alleviate dimensional disasters. The linear dimension reduction method PCA considers the global structure of the data; the manifold learning dimension reduction method LLE considers the local structure of the data. In this paper, by combining two dimensionality reduction methods, propose a new PCA_LLE algorithm,so that they complement each other"s advantages. Experiments on handwritten digital dataset,firt reduce the dimension of the dataset, and then use the K-nearest neighbor algorithm to classify the dimensionality-reduced data.Result show that the dimensionality reduction methods combining the two algorithms are improved compared with the original PCA and LLE algorithms. Moreover, the new algorithm PCA_LLE reduces the dimension reduction time of the new sample much more than the LLE algorithm.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭